This paper proposes an energy- and size-aware clustering (ESC) strategy for megaconstellation systems. The methodology constructs a hierarchical architecture with designated cluster heads and member satellites through role differentiation. Cluster heads serve as relay nodes, reducing the communication load on ground infrastructure by interacting with ground stations and other entities. This hierarchical approach enables efficient constellation control through interactions primarily with cluster heads. The ESC strategy begins with a cluster head election process, incorporating orphan handling management to select cluster heads, ensuring full coverage of the entire constellation. Then, a two-stage multi-objective optimization framework is used to deal with division of members to clusters and intra-cluster multi-hop connectivity configurations. This dual-phase approach optimizes energy consumption while maintaining balanced cluster sizes. Simulations on the OneWeb system validate the ESC strategy as a viable solution in that it achieves 55.9% energy consumption reduction compared to traditional discrete gradient optimization methods while maintaining more balanced cluster sizes.

Orbital altitude deviations and perturbations (e.g., Earth's J₂ effect, atmospheric drag) cause phase drift among multi-satellites, disrupting the overall configuration. This paper presents an optimization-based relative configuration maintenance strategy to maintain the orbital configuration of multiple satellites. We propose a bi-objective optimization framework, which independently optimizes the nominal point and the control input. Unlike conventional methods, the proposed approach systematically adjusts the expected positions of all satellites to minimize relative deviations during each control cycle. The strategy ensures that all satellites reach their assigned stationkeeping regions while minimizing the total energy consumption. In addition, phase errors among satellites are explicitly considered in the control input optimization, enabling a precise configuration maintenance. The strategy also accounts for a longterm orbital altitude stability. Simulation results under three distinct orbital altitude scenarios validate the effectiveness, stability, and robustness of the proposed approach in coordinating multisatellite systems under complex orbital conditions

The co-orbital multi-satellite system operating in low Earth orbit is prone to relative configuration instability due to orbital injection errors, J₂ perturbations, and atmospheric drag. However, maintaining a stable configuration is crucial for the normal operation of the mission. To address this issue, we propose a relative configuration maintenance strategy based on virtual satellites. Specifically, the strategy first establishes an optimization framework for selecting the virtual nominal point. This framework constructs a multi-objective function by integrating phase errors and satellite remaining energy, which is then solved using an optimization algorithm. Subsequently, control thresholds are set to ensure that phase errors remain within acceptable limits. Additionally, the strategy incorporates orbit altitude maintenance into the control law to ensure long-term stability. Simulation results validate the effectiveness of the proposed approach: compared to the baseline method, our strategy reduces the number of control maneuvers required, decreases the total phase control input, and achieves a more balanced energy distribution among the satellites.

Low-Earth-orbit (LEO) megaconstellations are transforming satellite communications from sparse, ground-controlled infrastructures into dense, dynamic, and increasingly autonomous space networks, while their global coverage capability is fundamentally enabled by large-scale symmetric orbital structures distributed across multiple planes and shells. As these systems expand to tens of thousands of satellites, maintaining such orbital symmetry under continuous perturbations, changing communication topologies, and varying onboard resources becomes a fundamental operational challenge. Future space systems must therefore manage, coordinate, and sustain large constellations for which their orbital configurations, communication topologies, and onboard resources vary continuously. Here, we review the management and configuration-maintenance problems of megaconstellations through a Starlink-informed perspective. We first summarize the multi-shell deployment architecture, satellite platform evolution, and dominant orbital perturbations that shape constellation behavior. We then examine hierarchical and cluster-based management strategies designed to reduce the burden on ground control and improve scalability. We further discuss in- and out-of-plane configuration maintenance. Finally, we identify open challenges in distributed autonomy, multi-shell coordination, dynamic topology management, and intelligent orbit control. This review highlights that the long-term viability of megaconstellations will depend not only on launch capacity and satellite manufacturing but also on scalable decision-making, autonomous coordination, and sustainable orbital operations.

Accurate prediction of battery remaining useful life (RUL) is essential for electric vehicles and energy storage systems. Existing methods face three challenges: insufficient modeling of temporal degradation structure, unstable predictions in the critical region near failure, and poor balance among multiple learning objectives. This paper proposes BiGRU-TACL, a deep multi-task learning framework for robust battery RUL prediction. First, unlike conventional contrastive learning with random temporal augmentation, the proposed temporal-aware contrastive learning (TACL) module constructs positive and negative sample pairs using cycle-number and RUL differences. This helps the model learn temporally ordered, stage-discriminative representations and a continuous degradation trajectory. Second, an online hard example mining (OHEM) mechanism, combined with a region-weighted regression loss and a boundary penalty, dynamically focuses on high-error samples and critical degradation regions instead of treating all samples equally. This improves prediction accuracy and stability near failure. Third, an adaptive weighting strategy based on homoscedastic uncertainty is introduced to automatically balance the RUL regression, health-stage classification, and contrastive learning objectives, avoiding manual weighting in conventional multi-task learning. Experiments on the NASA and CALCE datasets show that the proposed method achieves $ R^2 $ values of at least 0.96 and stage classification accuracies above 0.93, validating its effectiveness.